Pattern learning and recognition apparatus in a computer system
First Claim
1. Pattern recognition apparatus in a computer system, the apparatus comprising a feedback, self compensating network, the network receiving an input pattern provided to the computer system for recognizing and determining subpatterns thereof, the network simultaneously coding, through direct access of a content-addressable memory area, both the whole input pattern and various groupings of subpatterns in the input pattern, each said coding including a respective activity weight, the weight of a code indicating the probability of the input pattern being the grouping of subpatterns of that code, the subpatterns being independently recognizable patterns having exiting nodes in the memory area, the probability based upon spatial likeness between the grouping of subpatterns and the input pattern, with respect to previous and succeeding input patterns and past probabilities used for recognizing other input patterns, such that the code with a weight of highest probability indicates recognition of the input pattern as the grouping of subpatterns of the code.
0 Assignments
0 Petitions
Accused Products
Abstract
A masking field network F2, is characterized through systematic computer simulations serves or a content addressable memory. Masking field network F2 receives input patterns from an adaptive filter F1 →F2 that is activated by a prior processing level F1. The network F2 activates compressed recognition close that are predictive with respect to the activation patterns flickering across F1, and competitively inhibits, or masks, codes which are unpredictive with respect to the F1 patterns. The masking field can simultaneously detect multiple groupings within its input patterns and assign activation weights to the recognition codes for these groupings which are predictive with respect to the contextual information embedded within the patterns and the prior learning of the network. Automatic rescaling of sensitivity of the masking field as the overall size of an input pattern changes, allows stronger activation of a code for the whole F1 pattern than for its salient parts. Network F2 also exhibits adaptive sharpening such that repetition of a familiar F1 pattern can tune the adaptive filter to elicit a more focal spatial activation of its F2 recognition code than does an unfamiliar input pattern. The F2 recognition code also becomes less distributed when an input pattern contains more contextual information on which to base an unambiguous prediction of the F1 pattern being processed. Thus the masking field embodies a real-time code to process the predictive evidence contained within its input patterns. Such capabilities are useful in speech recognition, visual object recognition, and cognitive information processin
This invention was made with Government support under AFOSR-85-0149 awarded by the Air Force. The Government has certain rights in this invention.
123 Citations
37 Claims
- 1. Pattern recognition apparatus in a computer system, the apparatus comprising a feedback, self compensating network, the network receiving an input pattern provided to the computer system for recognizing and determining subpatterns thereof, the network simultaneously coding, through direct access of a content-addressable memory area, both the whole input pattern and various groupings of subpatterns in the input pattern, each said coding including a respective activity weight, the weight of a code indicating the probability of the input pattern being the grouping of subpatterns of that code, the subpatterns being independently recognizable patterns having exiting nodes in the memory area, the probability based upon spatial likeness between the grouping of subpatterns and the input pattern, with respect to previous and succeeding input patterns and past probabilities used for recognizing other input patterns, such that the code with a weight of highest probability indicates recognition of the input pattern as the grouping of subpatterns of the code.
-
6. A computerized pattern recognition and learning system comprising:
-
an adaptive filter that is activated by a source pattern and provides an input pattern indicative of the source pattern; a self-similar, automatically gain controlled cooperative-competitive nonlinear feedback masking field network having a content addressable memory responsive to the adaptive filter input pattern, the content addressable memory holding a plurality of list codes, and based on the input pattern, the network activating list codes that are predictive of the source pattern and competitively inhibiting list codes which are unpredictive of the source patterns based on different groupings of subpatterns in the input pattern, such that the masking field network provides the list code in the content addressable memory which is most predictive of the source pattern to indicate recognition of the source pattern and stores the source pattern with the list code in the content addressable memory to provide learning in the system. - View Dependent Claims (7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19)
-
-
20. A computerized recognition and learning system comprising:
-
a plurality of source nodes which embody a source pattern; a multiple scale, cooperative-competitive feedback masking field network of list nodes, the list nodes holding respective recognition codes of patterns; a plurality of adaptive filter paths which map the source nodes to list nodes in a manner which provides an input pattern to the network; the network receiving the input pattern from the adaptive filter paths and having content-addressable memory means responsive to the input pattern, the content-addressable memory means determining (i) list nodes which provide compressed recognition codes that are predictive of the source pattern, said list nodes being activated list nodes, and (ii) list nodes which provide compressed recognition codes that are unpredictive of the source pattern, said list nodes being masked list nodes, where at least one adaptive filter path is mapped from each source node to at least one activated list node, such that the masking field network provides the recognition code which is most predictive of the source pattern, based on different groupings of subpatterns in the input pattern, to indicate recognition of the source pattern, and stores the recognition code of the source pattern in a list node to provide learning in the system. - View Dependent Claims (21, 22, 23, 24, 25, 26, 27, 28, 29, 30)
-
-
31. A method of learning and recognizing patterns comprising the steps of:
-
providing an input pattern; simultaneously detecting multiple groupings of subpatterns within the input pattern, each grouping being associated with a predefined code accessible through a content-addressable memory; assigning respective weights to the codes for the detected multiple groupings; activating the codes associated with the detected groupings such that the codes of the groupings with their respective weights each provide a probability that the input pattern is the grouping associated with that code according to relative positions of subpatterns in the input pattern and codes used to recognize prior input patterns; and selecting the code with the weight of highest probability to indicate recognition of the input pattern as the grouping of subpatterns of the selected code. - View Dependent Claims (32, 33, 34, 35, 36, 37)
-
Specification